Ship Trajectory Online Prediction Based on BP Neural Network Algorithm

In view of most ship trajectory prediction methods based on dynamic or kinematics theory need set up ship motion model which is hard to define due to the real water condition, a new method based on Back Propagation (BP) neural network is proposed in this paper. By discussing the criterion factor affected ship position computation together with this model´s universality for any kind of ship, the model use ship course and speed as well as difference of longitude and latitude as input and output, respectively. At time n, using the lasted N training samples under the same time step train the network to capture the ship motion law, then predicted ship position at n+1. The experiment ship was in Changjiang River and the results demonstrated that the conventional method, like mercator computation, failed to get the correct results, but our method could capture the ship motion law within one second and has higher accuracy in prediction. With BP neural network´s excellent learning ability, this method can be used to any water condition which conventional method can´t deal with and moreover, avoided regular modeling process, which is especially suitable for the motion rules uncertain or unknown.